What Is Medallion Architecture? Bronze, Silver, and Gold Layers Explained

See how Bronze, Silver, and Gold refine ERP and HCM data into trusted metrics, reports, and AI-ready insight.

<p>Medallion architecture is a data design pattern that organizes data into Bronze, Silver, and Gold layers. Bronze preserves source-aligned data, Silver cleans it, and Gold prepares governed data models and metrics for business use. The pattern helps separate raw ingestion from trusted reporting, analytics, and AI consumption.</p> <p>Although <a href="https://www.databricks.com/blog/what-is-medallion-architecture">Databricks popularized medallion architecture</a>, the pattern is not exclusive to Databricks. It can be implemented across lakehouses, data lakes and cloud data warehouses, including environments built on <a href="https://learn.microsoft.com/en-us/fabric/onelake/onelake-medallion-lakehouse-architecture">Microsoft Fabric</a>, Snowflake and other modern data platforms.</p> <p>For business analytics, however, creating three technical layers is only the beginning. Data from sources like Oracle, Workday, SAP, UKG and other enterprise applications must retain enough structure, context and governance to support metrics that finance, HR and operational leaders can trust.</p> <h2>How do Medallion Architecture’s Bronze, Silver, and Gold layers work?</h2> <p>Each Medallion Architecture layer represents a different state of readiness. Data does not simply move from one storage location to another. It becomes progressively more reliable, consistent, and useful.</p> <table> <thead> <tr> <th><strong>Layer</strong></th> <th><strong>What it typically contains</strong></th> <th><strong>Required quality bar</strong></th> <th><strong>Primary consumers</strong></th> </tr> </thead> <tbody> <tr> <td>Bronze</td> <td>Raw or source-aligned records and ingestion metadata</td> <td>Complete and traceable representation of what arrived</td> <td>Data engineering, audit and troubleshooting teams</td> </tr> <tr> <td>Silver</td> <td>Cleansed, deduplicated and conformed entities</td> <td>Valid formats, keys, relationships and cross-system mappings</td> <td>Data engineering, analytics and data science teams</td> </tr> <tr> <td>Gold</td> <td>Business-aligned models, measures and aggregates</td> <td>Approved definitions, calculation logic, reporting grain and access rules</td> <td>Finance, HR, operations, BI and AI applications</td> </tr> </tbody> </table> <h3>Bronze layer: Preserve what the source recorded</h3> <p>The Bronze layer is the landing point for data from source systems. Its purpose is to preserve a source-aligned version of the data before substantial business transformations are applied.</p> <p>Depending on the source, Bronze might contain:</p> <ul> <li>Invoice headers and lines from an ERP</li> <li>Worker and employment records from an HCM system</li> <li>Opportunities and activities from a CRM</li> <li>Records from operational databases</li> <li>Data received through APIs or files</li> <li>Technical information about when and how the data arrived</li> </ul> <p>Bronze data is not necessarily accurate, deduplicated or ready for reporting. That is intentional. Its value lies in retaining a traceable foundation that teams can refer to when investigating an issue or applying revised transformation logic.</p> <p>In SplashBI, <a href="/enterprise-intelligence-platform/data-pipeline">ready-made data pipelines</a> can extract data from sources such as Oracle, Workday, Salesforce and UKG and deliver it to a selected warehouse, BI tool or AI environment.</p> <h3>Silver layer: Create reliable enterprise entities</h3> <p>The Silver layer is where data is cleaned, standardized and connected.</p> <p>This might include:</p> <ul> <li>Removing duplicate records</li> <li>Standardizing date and currency formats</li> <li>Resolving missing or invalid values</li> <li>Mapping different source identifiers to a common business entity</li> <li>Aligning organizational structures and hierarchies</li> <li>Connecting related records from multiple systems</li> <li>Applying defined validation rules</li> </ul> <p>The most important job of Silver is not making columns look tidy. It is creating reliable representations of business entities such as a supplier, invoice, worker, customer, department, or account.</p> <p>For example, the same supplier may appear under different names or identifiers in an ERP, procurement application, and payment system. Until those records are matched correctly, a downstream supplier-spend report may be incomplete or misleading.</p> <p>Silver therefore provides the bridge between source-specific structures and the way the organization understands its operations.</p> <h3>Gold layer: Define the business answer</h3> <p>The Gold layer organizes data for specific business uses. It may contain reporting models, dimensional models, approved measures, aggregates, and domain-specific data products.</p> <p>Typical Gold-layer outputs include:</p> <ul> <li>Accounts payable aging</li> <li>Budget-versus-actual reporting</li> <li>Workforce headcount</li> <li>Employee turnover and attrition</li> <li>Sales pipeline analysis</li> <li>Inventory and production metrics</li> <li>Executive dashboards</li> <li>Data prepared for natural-language querying</li> </ul> <p>Gold data should answer more than “Is this record technically valid?” It should answer questions such as:</p> <ul> <li>What exactly counts as an active worker?</li> <li>Which date determines an invoice’s aging bucket?</li> <li>How is revenue recognized?</li> <li>Which organizational hierarchy applies?</li> <li>What information is this user permitted to see?</li> <li>Which calculation is approved for management reporting?</li> </ul> <p>A semantic layer may sit within Gold or be maintained as a separate layer above it. There is no universal naming rule. What matters is that business definitions, relationships, hierarchies, and access rules are explicit and consistently applied.</p> <h2>Why use medallion architecture?</h2> <p>Raw data from an ERP, HCM or CRM application is rarely ready to enter a board report or answer an executive question. It may contain duplicates, application-specific codes, different calendars, inconsistent identifiers, or relationships that only make sense within the source system.</p> <p>Medallion architecture separates the work of preserving, preparing and serving that data.</p> <h3>Medallion Architecture isolates raw data from business consumption</h3> <p>Business users and analytics applications do not need to navigate raw transactional structures. They can work with models designed for their questions while the source-aligned records remain available for validation and troubleshooting.</p> <h3>Medallion Architecture introduces quality progressively</h3> <p>Different quality checks belong at different stages. Bronze can be checked for completeness and traceability. Silver can be checked for valid keys, standardized formats, and reliable relationships. Gold can be reconciled against approved metrics and reporting rules.</p> <h3>Medallion Architecture supports reuse</h3> <p>A conformed supplier, worker or account entity can support multiple reports and use cases. Teams do not have to clean and interpret the same source records separately for every dashboard.</p> <h3>Medallion Architecture makes problems easier to locate</h3> <p>If a report contains the wrong number, teams can investigate the calculation in Gold, the conformed entities in Silver and the original source-aligned records in Bronze. Clear boundaries make it easier to identify where an error entered the process.</p> <h3>Medallion Architecture gives different users the appropriate data</h3> <p>Data engineers may need source-level detail. Analysts may need conformed but granular data. Business leaders need governed measures and reports. Medallion architecture can provide these different access points without treating every dataset as equally ready for every purpose.</p> <p>However, the architecture does not guarantee data quality merely because three layers exist. The quality of the result still depends on clear rules, validation, monitoring, ownership and governance.</p> <h2>How does medallion architecture improve data quality?</h2> <p>Medallion architecture creates defined points at which different types of quality controls can be applied.</p> <table> <thead> <tr> <th><strong>Layer</strong></th> <th><strong>Examples of relevant quality checks</strong></th> </tr> </thead> <tbody> <tr> <td>Bronze</td> <td>Expected records or files arrived, required source fields are present, ingestion is traceable</td> </tr> <tr> <td>Silver</td> <td>Duplicate records are addressed, identifiers are valid, formats are standardized, relationships are reliable</td> </tr> <tr> <td>Gold</td> <td>Metrics reconcile with approved sources, calculation rules are documented, reporting grain is correct, access is appropriate</td> </tr> </tbody> </table> <p>These checks should reflect the organization’s actual business requirements. A field can be technically valid and still be unsuitable for a particular report.</p> <p>For example, an invoice date may be populated correctly but may not be the date that finance uses to determine an aging bucket. Similarly, an employee record may be complete but still require effective-date logic before it can be included correctly in historical headcount.</p> <p>The goal is therefore not simply cleaner data. It is data that is fit for a defined business purpose.</p> <h2>From ERP and HCM data to a trusted business metric</h2> <p>The clearest way to understand medallion architecture is to follow a business question through the three layers.</p> <h3>Finance example: From invoices to AP aging</h3> <p>Consider the question:</p> <p>What is our current accounts payable exposure by supplier and aging bucket?</p> <p>A medallion implementation could process the required data as follows:</p> <p><strong>Bronze</strong></p> <p>Invoice headers, invoice lines, supplier records, payment schedules, and related accounting data arrive from the source system in source-aligned structures.</p> <p><strong>Silver</strong></p> <p>Supplier identities are matched, invoice statuses are standardized, currencies and accounting periods are aligned, and related invoice and payment records are connected.</p> <p><strong>Gold</strong></p> <p>Open balances are assigned to approved aging buckets. The resulting model organizes the data by supplier, legal entity, business unit, reporting date, and other governed dimensions.</p> <p><strong>Business consumption</strong></p> <p>Finance teams access AP aging through a report, dashboard, spreadsheet workflow or natural-language question.</p> <p>SplashBI’s <a href="/products/analytics/financial-analytics">Financial Analytics</a> provides reporting across Oracle EBS, Fusion Cloud and connected ledgers, including finance domains such as AP, AR, cash and the close process.</p> <p>A metric such as days payable outstanding would require additional data and an approved calculation method. It should not be treated as a direct transformation of invoice records alone.</p> <h3>HR example: From worker events to attrition</h3> <p>Now consider:</p> <p>What was voluntary attrition last quarter, and which departments experienced the largest increase?</p> <p><strong>Bronze</strong></p> <p>Worker, job, position, organization and employment-event data arrives from the relevant HCM systems.</p> <p><strong>Silver</strong></p> <p>Worker identities, effective dates, employment statuses, and organizational hierarchies are standardized. Records from different sources can be conformed where required.</p> <p><strong>Gold</strong></p> <p>The organization’s approved definitions of headcount, exits, voluntary attrition, reporting period and excluded populations are applied.</p> <p><strong>Business consumption</strong></p> <p>HR leaders receive consistent attrition results through dashboards, reports, or conversational analytics.</p> <p>SplashBI’s <a href="/products/analytics/people-analytics">People Analytics</a> brings workforce information into a unified analytics environment, while its <a href="/solutions/workforce-intelligence">Workforce Intelligence</a> offering connects governed HCM data with analytics and AI experiences.</p> <p>These examples illustrate an important distinction: business-ready data is not simply clean data. It is data interpreted using agreed business definitions.</p> <h2>Where does governance belong in medallion architecture?</h2> <p>Governance should not appear only after data reaches Gold.</p> <p>If authentication, lineage, access rules and business definitions are added only at the reporting stage, different tools may apply them differently. A dashboard, spreadsheet, and AI application could then produce different views of what should be the same metric.</p> <p>Governance across a medallion architecture can include:</p> <ul> <li>Authentication and identity</li> <li>Role-based access</li> <li>Row-level and column-level controls</li> <li>Source and transformation lineage</li> <li>Business definitions</li> <li>Data ownership</li> <li>Audit and monitoring records</li> <li>Rules governing how data may be consumed</li> </ul> <p>SplashBI describes its approach as one governed foundation across the data journey. Its <a href="/enterprise-intelligence-platform/connectors">enterprise data connectors</a> include role-based access, row-level security, audit logs and enterprise identity support.</p> <p>This becomes especially important when data is made available to AI. According to the <a href="/company/press/splashbi-advances-tahoe-6-2">SplashBI Tahoe 6.2 announcement</a>, SplashBI reuses existing BI security controls, including role-based and row-level controls, for AI-generated insights rather than requiring a separate AI security model.</p> <p>The governing principle is simple: Governance is not a fourth layer. It must surround Bronze, Silver and Gold.</p> <h2>Is medallion architecture only for Databricks?</h2> <p>No. Medallion architecture is a design pattern, not a proprietary product requirement.</p> <p>Databricks popularized the Bronze, Silver and Gold terminology within the lakehouse ecosystem. However, equivalent layers can be implemented on other platforms.</p> <table> <thead> <tr> <th><strong>Technology</strong></th> <th><strong>Its role in a medallion architecture</strong></th> </tr> </thead> <tbody> <tr> <td>Databricks and Delta Lake</td> <td>A prominent lakehouse implementation associated with the Bronze, Silver and Gold terminology</td> </tr> <tr> <td>Microsoft Fabric and OneLake</td> <td>Supports medallion implementation through separate lakehouses, warehouses or schemas</td> </tr> <tr> <td>Snowflake</td> <td>Can store and transform raw, conformed and curated datasets using schemas, tables and Snowflake data-engineering capabilities</td> </tr> <tr> <td>dbt</td> <td>Can manage transformations and tests between layers, but is not the underlying storage architecture</td> </tr> <tr> <td>Oracle, Workday, SAP and UKG</td> <td>Usually act as enterprise source applications in this scenario</td> </tr> <tr> <td>SplashBI</td> <td>Connects enterprise sources, moves and models data, and delivers it to reporting, analytics and AI experiences</td> </tr> </tbody> </table> <p>Microsoft describes medallion architecture as its recommended lakehouse design approach for Fabric. Snowflake also demonstrates the pattern in its <a href="https://www.snowflake.com/en/developers/guides/building-retail-analytics-de-pipeline/">data-engineering implementation guides</a>.</p> <p>SplashBI is designed to work with existing enterprise investments. Its <a href="/enterprise-intelligence-platform">Enterprise Intelligence Platform</a> connects source systems, moves data through Bronze, Silver and Gold layers, applies business models and delivers analytics through dashboards, natural-language queries and generated narratives.</p> <figure><img src="/__l5e/assets-v1/e602ddf7-253d-4d9b-961b-91c173818815/medallion-architecture.png" alt="SplashBI's medallion architecture for enterprise analytics" loading="lazy" /></figure> <h2>Is medallion architecture ETL or ELT?</h2> <p>Medallion architecture is neither ETL nor ELT. It describes how data is organized as its quality and business readiness increase. ETL and ELT describe when transformation occurs relative to loading.</p> <p>Many medallion implementations are ELT-like:</p> <ol> <li>Data is extracted from its source.</li> <li>It is loaded into Bronze with minimal transformation.</li> <li>It is transformed as it moves into Silver and Gold.</li> </ol> <p>However, an organization may still use ETL for particular sources or workloads. The choice does not determine whether the overall architecture qualifies as medallion.</p> <h2>Medallion architecture versus related data concepts</h2> <p>Several terms commonly appear alongside medallion architecture, but they are not interchangeable.</p> <table> <thead> <tr> <th><strong>Concept</strong></th> <th><strong>Relationship to medallion architecture</strong></th> </tr> </thead> <tbody> <tr> <td>Data lakehouse</td> <td>A data platform architecture in which medallion is frequently implemented</td> </tr> <tr> <td>ETL or ELT</td> <td>The method used to extract, load and transform data</td> </tr> <tr> <td>Star schema</td> <td>A reporting-oriented data model that may be used in Gold</td> </tr> <tr> <td>Semantic layer</td> <td>The layer that defines business meaning, measures and relationships; it may sit within or above Gold</td> </tr> <tr> <td>Data mesh</td> <td>An ownership and operating model that can use medallion within individual data domains</td> </tr> <tr> <td>Delta Lake</td> <td>A storage technology commonly associated with medallion implementations</td> </tr> </tbody> </table> <p>A lakehouse provides an environment for storing and processing data. Medallion architecture organizes data by its level of refinement. A star schema structures selected data for efficient analytical use. These concepts can work together rather than replacing one another.</p> <h2>How SplashBI applies medallion architecture to business analytics</h2> <p>SplashBI applies the Bronze, Silver and Gold pattern within a broader enterprise analytics journey:</p> <ol> <li><strong>Connect:</strong> Bring data in from ERP, HCM, CRM, databases, cloud platforms, files and APIs.</li> <li><strong>Pipe:</strong> Move and transform data through progressively refined layers.</li> <li><strong>Model:</strong> Apply business models for finance, HR, sales and operations, or use customer-defined models.</li> <li><strong>Analyze:</strong> Deliver data through reports, dashboards, self-service analytics and natural-language experiences.</li> </ol> <p>These stages are illustrated on the <a href="/enterprise-intelligence-platform">SplashBI Enterprise Intelligence Platform</a> page.</p> <p>The <a href="/company/press/splashbi-advances-tahoe-6-2">Tahoe 6.2 release</a> introduced medallion architecture support across SplashBI’s proprietary Workforce and Financial Analytics solutions. The same release also documents support for customer data warehouses, customer-defined data models, bring-your-own-model options and customer-hosted cloud environments.</p> <p>For organizations that already have a warehouse strategy, <a href="/enterprise-intelligence-platform/data-pipeline">SplashBI Data Pipeline</a> can deliver data to existing warehouse, BI and AI destinations. For business consumption, <a href="/products/splashai">SplashAI</a> provides conversational analytics grounded in finance and HCM business contexts.</p> <p>The purpose is not to create Bronze, Silver and Gold for their own sake. It is to give every downstream experience a more consistent, governed data foundation.</p> <h2>From raw records to answers the business can use</h2> <p>Medallion architecture gives data teams a useful structure for separating source ingestion, data preparation and business consumption.</p> <p>But the names of the layers are less important than the questions they answer:</p> <ul> <li>Does Bronze preserve what arrived from the source?</li> <li>Does Silver create reliable and reusable enterprise entities?</li> <li>Does Gold apply the definitions needed for business decisions?</li> <li>Does governance remain intact across every layer and consumption channel?</li> </ul> <p>When those conditions are met, medallion architecture can connect complex enterprise systems to reports, dashboards and AI experiences that business teams can use with greater confidence.</p> <p><strong><a href="/schedule-demo">Schedule a SplashBI demo</a> to explore how a governed Bronze, Silver and Gold foundation can support your enterprise reporting, analytics and AI strategy.</strong></p>

FAQ

What is medallion architecture?

Medallion architecture is a data design pattern that separates data into Bronze, Silver and Gold layers. Data moves from a raw or source-aligned state through cleansing and conformance before it is organized for reporting, analytics, machine learning or AI.

What are the three layers of medallion architecture?

Bronze contains raw or source-aligned data. Silver contains cleansed and conformed data. Gold contains business-ready models, measures and aggregates designed for specific analytical uses.

What is the Silver layer in medallion architecture?

The Silver layer cleans, standardizes, and connects data from Bronze. It may resolve duplicate records, align identifiers, standardize formats, and create reliable business entities that can be reused across analytical use cases.

Why use medallion architecture?

Medallion architecture separates raw ingestion from business consumption, introduces data-quality checks progressively, and makes it easier to trace, reuse and troubleshoot data.

Is medallion architecture specific to Databricks?

No. Databricks popularized the Bronze, Silver and Gold terminology, but medallion architecture is a vendor-agnostic design pattern. It can be implemented on platforms such as Microsoft Fabric, Snowflake, and other lakehouse or cloud warehouse environments.

Does Snowflake support medallion architecture?

Yes. Organizations can represent Bronze, Silver and Gold through separate Snowflake databases, schemas or tables and use Snowflake capabilities to ingest, transform and serve the data.

Is medallion architecture ETL or ELT?

It is neither. Medallion architecture describes how data is organized by its level of refinement. Many implementations are ELT-like because data is loaded into Bronze before being transformed into Silver and Gold.

Is the Gold layer in medallion architecture the same as a semantic layer?

Not always. Some organizations include governed measures and business meaning within Gold. Others maintain a separate semantic layer above it. The boundary depends on the implementation.